refactor(header): replace IsEmptyBlock with fast IsZero short-circuit

[ValentaTomas]

IsEmptyBlock was the per-block hot path of DiffMetadataBuilder.Process — one call per block during the rootfs scan in DirectProvider.exportToDiff. It paid a full SIMD memcmp (4 KiB or, on hugepages, 2 MiB) before it could short-circuit, and required two static zero buffers (EmptyBlock, EmptyHugePage) to compare against.

Replace it with a small IsZero helper that uses QEMU's buffer_is_zero trick: sample three bytes (first, last, middle) so most non-zero blocks reject from a single cache line. The fallback uses bytes.Equal on b shifted by one (b[1:] == b[:n-1]), which dispatches to the runtime's SIMD memequal on amd64/arm64 — same speed as the old comparison against a static buffer when the short-circuit doesn't fire.

DiffMetadataBuilder.Process calls IsZero directly (the size validation IsEmptyBlock did was unreachable in practice). EmptyBlock is removed; EmptyHugePage stays — build.go uses it as the literal zero buffer for uuid.Nil reads in Build.ReadAt.

Pre-factored from #2546 so it can land on its own.

[cursor]

PR Summary

Medium Risk
Touches the hot-path logic that decides whether a block is treated as empty vs dirty; any false positives would silently drop data from diffs. The new heuristic is simple but should be validated across all block sizes and architectures to avoid edge-case misclassification.

Overview
Replaces per-block zero detection in DiffMetadataBuilder.Process by removing IsEmptyBlock and introducing a faster IsZero check that short-circuits on sampled bytes before a SIMD-backed equality fallback, and adds unit tests covering sampled and buried non-zero cases.

^{Reviewed by Cursor Bugbot for commit 88dfb98. Bugbot is set up for automated code reviews on this repo. Configure here.}

[claude]

Small, well-tested refactor of a leaf hot-path utility — only finding is a doc-comment nit, safe to land.

Extended reasoning...

Overview

This PR refactors header.IsEmptyBlock (per-block hot path of the rootfs diff scan) in packages/shared/pkg/storage/header/diff.go plus adds a new diff_test.go. The substantive changes are: (1) a new IsZero helper that samples 3 bytes (first/last/middle) before falling back to a SIMD-accelerated self-equality compare, (2) deletion of the now-unused EmptyBlock global, (3) explicit length validation in IsEmptyBlock, (4) updated doc comments. About 40 lines net added, all in a self-contained leaf utility.

Security risks

None. This is pure CPU-side logic over already-loaded buffers — no auth, no I/O, no untrusted input parsing. The new length check actually tightens the API surface (callers can no longer silently pass a wrong-shape buffer). The deletion of EmptyBlock is verified safe: grep shows no remaining references after the patch.

Level of scrutiny

Low. Tight scope, behavior-preserving by intent, and the correctness argument for IsZero is straightforward to verify: for n>3, the 3-byte sample forces b[0]==0, and bytes.Equal(b[:n-1], b[1:]) then transitively requires all adjacent bytes to be equal, so all bytes must be zero. For n≤3, the sample alone covers every index. The new tests exercise nil, len≤3, full-block zero, and non-zero bytes hidden between sample positions. Length-mismatch and unsupported-block-size paths are also tested.

Other factors

The single bug finding is a doc-comment nit: the new comment for EmptyHugePage references Build.ReadAt, but the actual consumer is File.Slice at build.go:107 (verified via grep — no Build.ReadAt exists in this repo). Doc-only, no runtime impact, doesn't justify blocking. The PR description carries the same minor inaccuracy. Otherwise the change is mechanical and well-motivated.

[ValentaTomas]

Trimmed IsZero doc to one paragraph and dropped the trivial test cases — kept just the cases that exercise the 3-byte sample, the SIMD fallback path, and the byte-just-before-the-last-sample edge case. Also fixed the EmptyHugePage callsite reference to (*build.File).Slice (was Build.ReadAt per Claude review).